search

AI4S2S / s2spy

A high-level python package integrating expert knowledge and artificial intelligence to boost (sub) seasonal forecasting
https://ai4s2s.readthedocs.io/
Apache License 2.0
20 stars 7 forks source link

Add "gap_lag" and "heterogeneous time aggregation" (RGDR relevant) features/calendar #47

Closed geek-yang closed 1 year ago

geek-yang commented 2 years ago

Two new features relevant for RGDR need to be accommodated by the time.py module later:

(1) Lag as gap For RGDR, we might want to define lags in a more precise manner. This gives more flexibility to define the lead-time of the forecast.

image

(2) Heterogeneous time aggregation The further you go back in time, the more you might want to focus on low-frequency variability.

image

Whether these two features need to be added as features or new calendar types need to be discussed. But for now, this task is not prioritized. This issue ensures that the request is well documented.

This issue is also relevant to issue AI4S2S/lilio#9.

Peter9192 commented 2 years ago

Just to articulate: with lag_as_gap you would only need the single interval that corresponds to that lag, right? Not the 10 lags before that.

By definition, the period has a start/center/end. So if I understand you correctly, you want to precisely align the end of the precursor to the start of the target (which is currently not possible, since you are limited to lags ending on days that are exactly offset n days from the end of the target period, where n is the number of days, i.e. the inverse frequency). Right?

semvijverberg commented 2 years ago

Yes that is correct @Peter9192. As a heads up, this feature will be needed for the upcoming Lorentz workshop. I will share more the details on that soon.

BSchilperoort commented 1 year ago

Implemented by AI4S2S/s2spy#113